Supermicro has announced 12 new server platforms built around Intel's latest Xeon 6+ processors. No GPUs. No AI training workloads. No mentions of Jensen Huang.
This is a story about the other kind of data centre computing, the kind that keeps the internet running while everyone else argues about who has the most Nvidia chips.
What the servers actually do
Strip away the jargon and the announcement is straightforward. These are machines designed to pack as much processing power as possible into as little space and energy as possible.
The top-end configurations fit 576 processor cores into a single server. For context, a typical laptop has between four and 16 cores. These machines are built for cloud providers and large enterprises that need to run enormous numbers of tasks simultaneously, things like web hosting, databases, video streaming and the general background plumbing of the internet.
The key metric is performance per watt, meaning how much work the server does relative to the electricity it consumes. Data centres are increasingly constrained by power supply, so a server that does more work on the same energy bill is worth paying for.
Why Intel matters here
Intel has spent the past three years watching Nvidia and AMD dominate the AI conversation. Its share of the data centre GPU market is negligible. But the CPU market, the processors that handle everything that is not AI training, is still Intel's territory.
The Xeon 6+ chips in these servers offer double the core count of the previous generation, faster memory support and significantly more cache, which is the processor's short-term memory for frequently used data.
None of that is glamorous. All of it matters for the companies paying the electricity bills.
What Supermicro is really doing
Supermicro made its name as the company that could ship AI servers faster than anyone else during the GPU boom. The stock went from obscurity to a market capitalisation above $60bn on the back of that reputation.
But AI servers are a cyclical, margin-thin business dependent on Nvidia allocation. Expanding the product line into high-density, energy-efficient CPU servers is a hedge, a way to diversify revenue into workloads that are less dependent on a single supplier and less vulnerable to the boom-and-bust cycle of AI capex.
It is not the exciting story. It might be the smarter one.
